How does Scene/SceneRoot's World interact with the main world?

[steplee]

I'm observing what appears to be a frustum culling bug on an imported gltf scene. This lead me to digging into some internals.

I wanted to add ShowAabbGizmo but it doesn't work. I suspect that is because the SceneRoot component does not have a Mesh3d so an AABB is never added to it. That's where I wanted to dive in and see how I could interact with the Scene's inner world, or even easier: pull all of it's contents out into the main world. That might even be what's being done when it is spawned but after an hour of reading through source code I cannot tell.

So that lead me to a few general questions out of curiosity:

• Does a Scene store it's own World recursively as a child in the main World?
• If so, where does bevy_render and/or the extraction process take this into account? And if not, how does it work? Just spawning all things in the Scene's world in the main one, and using Parent/Children links?

And more specific questions about my particular suspected issue:

• How can I simply add the ShowAabbGizmo to a mesh within a SceneRoot?
• Am I barking up the wrong tree with GltfScenes? This asset contains animations, so I think I must use the Scene and not e.g. the sole GltfPrimitive converted to a Mesh3d (the fox examples use the full scenes, which is all I have to go off of for idiomacy).

[steplee]

Answered my own questions after a night's rest 😜 :

1. A Scene does not store it's own inner World. It just adds all entities in the scene to the main world directly, and keeps track of identifiers in an internal structure so it knows how to remove them when despawned. This is most clearly demonstrated in Scene::write_to_world_with + SceneSpawner::spawn, the latter whose World argument is passed from scene_spawner_system, which being a system, means that it's World argument is from the main world.
2. The question about bevy_render is not relevant because there is only the one main world.
3. A ShowAabbGizmo could be added to every descendant Entity of a scene that contains an AABB like so (but should also take care to do this only once...):

fn debug_scene(
        q_toplevel_parent: Query<(&bevy_scene::SceneInstance, &Children, Entity)>,
        q_any_parent: Query<(&Children, Entity)>,
        q_has_aabb: Query<Entity, With<bevy_render::primitives::Aabb>>,
        mut commands: Commands,
    ) {
    if let Ok((_, children, parent_ent)) = q_toplevel_parent.get_single() {

        let mut stack = Vec::new();
        let mut descendents = Vec::new();

        for &child_ent in children.iter() {
            stack.push(child_ent);

            while stack.len() > 0 {
                let it = stack.pop().unwrap();
                descendents.push(it);
                if let Ok((cs,p)) = q_any_parent.get(it) {
                    for c in cs.iter() { stack.push(*c) };
                }
            }
        }

        info!("total descendents: {}", descendents.len());
        for &d in descendents.iter() {
            if let Ok(ent) = q_has_aabb.get(d) {
                info!("descendent has aabb");
                commands.entity(ent).insert(ShowAabbGizmo::default());
            }
        }
    }
}

This helped me confirm there is something wrong with the AABB calculation but the frustum culling is correct!